The Signal Environment of a Wearable ECG Patch Is Hostile by Definition
A hospital-grade 12-lead ECG is recorded under controlled conditions: patient lying still, limb leads correctly placed, room-temperature gel ensuring electrode contact, a recording time of 10 seconds. The entire clinical setup is optimized to produce clean signal.
A wearable ambulatory ECG patch operates in the opposite environment. The patient is moving, sleeping, showering (if the patch is waterproof-rated), sweating during exercise, and wearing the electrode against skin that changes in hydration and temperature over 7 to 14 days. The signal chain from skin to recorder must handle all of that for the entire monitoring period, without clinical supervision. Signal quality on day 14 is almost always meaningfully worse than on day 1. The question for any automated analysis system is how to handle that degradation without producing classification errors that reach clinical review disguised as high-confidence findings.
The Four Main Noise Sources in Ambulatory ECG Patch Data
Motion Artifact
Electromyographic (EMG) interference from skeletal muscle activity is the most common source of noise in ambulatory recordings. Walking, lifting, and even shallow breathing can introduce EMG signals that partially overlay the ECG waveform in the same frequency band as the QRS complex. When EMG power density is high enough, the R-wave peak becomes difficult to isolate from the noise floor, and RR interval calculations become unreliable.
The challenge is that some motion artifact patterns are rhythmic. Gait produces a periodic signal. If a classifier is not explicitly designed to distinguish cardiac from locomotion-derived periodicity, it can assign an incorrect rhythm label to a noise-contaminated segment. This is one of the reasons that frequency-domain filtering alone is insufficient. Motion artifact rejection requires a combination of spectral and amplitude-based methods, and the threshold parameters need to be validated against real ambulatory recordings, not bench-test data.
Electrode-Skin Interface Degradation
Patch electrodes are in continuous contact with skin for the full monitoring period. As adhesive hydrogel dries out and skin cells shed beneath the electrode, contact impedance increases. Higher impedance means the bioelectric signal is attenuated relative to environmental electrical noise. The result is a gradually declining signal-to-noise ratio across the monitoring window.
Electrode lift, partial detachment of one edge of the patch, is a related but distinct problem. Lift typically produces an abrupt amplitude change followed by intermittent signal loss as the electrode alternately makes and loses contact. These transitions can look superficially like atrioventricular block or extreme bradycardia to a classifier operating on raw RR intervals. Detecting electrode status transitions in real time and flagging the affected data segments is a preprocessing step that has to happen before arrhythmia classification, not after.
Baseline Wander
Low-frequency drift in the ECG baseline is primarily caused by respiratory movement and patient position changes. As the thorax expands and contracts with each breath, the electrical axis of the heart shifts slightly, and the skin-electrode distance changes fractionally. These changes produce a slow sinusoidal variation in the recorded signal baseline at respiratory frequency (typically 0.15 to 0.4 Hz).
Baseline wander is usually removable with a high-pass filter, but the filter cutoff needs to be chosen carefully. Cut too low and baseline drift persists. Cut too high and the ST segment morphology is distorted, which affects any analysis that depends on ST level. For rhythm classification focused on rate and RR interval analysis, baseline wander is a manageable nuisance. For analysis that includes ST morphology (ischemia detection or bundle branch block characterization), wander correction is a critical preprocessing step that directly affects clinical accuracy.
External Electrical Interference
Power-line interference at 50 or 60 Hz and its harmonics is the fourth common noise source. Modern ambulatory patch hardware typically includes notch filtering for mains frequency, so this is less problematic than it was with older Holter recorders. However, high electromagnetic field environments (near MRI suites, certain industrial workplaces, or in rare cases near high-voltage infrastructure) can still introduce interference that notch filtering does not fully suppress. For most outpatient ambulatory monitoring, power-line interference is a known and well-managed problem, not an active concern.
How Automated Analysis Handles Noise Before Classification
The preprocessing pipeline that prepares raw patch data for arrhythmia classification is where most quality-related accuracy differences between systems actually live. Detection algorithms often perform similarly on clean data. The differences emerge in degraded signal conditions.
ElectroKare's processing pipeline applies noise characterization before passing any segment to the rhythm classifier. Each segment of the recording receives a signal quality index (SQI) computed from spectral and amplitude features. Segments below a defined SQI threshold are treated differently from clean segments. They are not simply discarded, which would create gaps in the monitoring record, and they are not analyzed as if they were clean data, which would produce high-confidence false classifications. Instead, they are flagged and the classification output for those segments is assigned a reduced confidence score.
The clinical consequence of this approach is that low-quality segments produce alerts with explicit confidence qualification rather than unqualified arrhythmia labels. A finding that emerges from a heavily artifact-contaminated segment should reach cardiologist review with that context attached. The cardiologist reviewing the ECG strip can see the signal quality directly and make the determination. ElectroKare surfaces the SQI context with every alert where quality was a factor, because the cardiologist's clinical judgment is the appropriate filter, not an automated confidence cutoff that silently buries the finding.
When Should a Clinician Be Notified About Signal Quality?
Some cardiologists ask whether low-quality data should simply be excluded from the analysis. We think the more useful framing is: what clinical action is appropriate for each quality scenario?
For a brief artifact burst lasting 30 to 60 seconds in an otherwise clean 14-day recording, exclusion and silent imputation are appropriate. For extended low-quality periods spanning multiple hours, the cardiologist should know. Not because the analysis is wrong, but because a multi-hour gap in the monitoring record during a period of potential clinical significance (days 1 through 3 post-discharge, for example) changes the interpretation of any negative finding from "no arrhythmia detected" to "no arrhythmia detected in the analyzable portion of this recording."
The distinction matters for how the cardiologist communicates findings to the patient and whether repeat monitoring should be considered. A monitoring report that shows good overall recording quality gives clinically meaningful confidence to a negative result. A report where 30 percent of the recording period was flagged as low-quality warrants a more qualified interpretation.
Our monitoring notes in the alert queue include a recording quality summary for each completed monitoring period. When a meaningful proportion of a patient's data was artifact-affected, that is surfaced as a monitoring note before the rhythm findings are reviewed, not buried in a technical appendix at the end of a PDF report.
What Practices Can Do to Improve Patch Data Quality
Clinical staff have more influence over patch data quality than most practices recognize. The two highest-impact interventions are both procedural, not technical.
First, skin preparation before patch application. Adequate degreasing of the application site reduces initial contact impedance and extends the useful life of the adhesive hydrogel. Removing residual body lotion, cleaning with an alcohol wipe, and allowing the skin to dry completely before application takes two minutes and measurably affects the signal quality of the entire monitoring period.
Second, patient education about positioning and activity. Patients who understand why they should avoid vigorous upper-body exercise during the first 24 hours (until adhesive fully sets), why sleeping on the side opposite the patch location helps reduce baseline wander during overnight recordings, and why any unusual sensation or visible edge-lift should be reported early, tend to produce recordings with better analyzable yield. This is not about restricting activity. Standard ambulatory monitoring should not substantially alter a patient's daily life. It is about a few specific behaviors during the initial adherence period that have disproportionate effects on recording quality.
Noise rejection algorithms can compensate for some degraded signal conditions. They cannot replace a well-applied patch on properly prepared skin. Both matter, and the clinical team controls one of them entirely.